Urban grassland restoration: which plant traits make desired species successful colonizers?

Authors


leonie.fischer@tu-berlin.de

Abstract

Questions

Which plant traits characterize successful and failed target species in urban grassland restoration? Do traits of successful target species differ from those of resident species? How do plant traits relate to environmental constraints?

Location

In-situ experimental sites on wastelands in shrinking residential areas in Berlin, Germany.

Methods

We established grassland restoration treatments and explored plant traits of successful and failed target species (plant height, specific leaf area, seed mass, seed shape, seed longevity index, CSR strategy type, plant life form). To shed light on mechanisms that shape restoration success, we also analysed the same traits of species originating from the soil seed bank and species immigrating from the surroundings. We compared both trait sets to those of resident species. With RLQ analyses we related the trait data to abundance data of species and to variables describing the environmental setting of the sites.

Results

In the third year after treatment, several plant traits differed between the successful or failed target species and the resident vegetation, e.g. successful target species tended to be as tall as resident species, whereas failed target species were smaller, suggesting insufficient competitive ability of the latter. Species that successfully recruited from the soil seed bank were taller than resident species. Small specific leaf area was important for the establishment success of target species. Trait composition of the species assemblage clearly related to the environmental setting: mean specific leaf area and the proportion of annuals increased and the proportion of C-strategists decreased with increasing human-mediated impacts on the restoration sites.

Conclusions

Our results reveal clear trait differences between successful and failed target species in grassland restoration on urban wasteland sites, demonstrating that high competitive ability is crucial for success in target species. Grassland species that are successfully integrated into urban wasteland vegetation may thus fill well-defined vacant niches, while resembling the traits of the resident species in other ways. Our results may allow generalizations and transfer to similar urban settings, as the analysed trait states were assessed as relative values compared to resident species.

Nomenclature
Rothmaler

(2005)

Introduction

In restoration projects that translocate desired species with the goal of developing target communities (Seddon 2010), competition among species of the resident vegetation and target species can strongly shape species establishment in grassland restoration (Zobel et al. 1996; Pywell et al. 2002; Zeiter et al. 2006; Tschöpe & Tielbörger 2010). A key question for the successful establishment of grassland species is thus how species cope with competitive pressure. Generally, species can compete against other species through their own competitive ability (Grime 2006), by filling vacant niches (Funk et al. 2008) or through facilitative effects (Leger & Espeland 2010). Underlying traits of successfully established target species may therefore either resemble the traits of resident species or differ greatly from them (Roberts et al. 2010). Using species traits to guide the choice of target species in grassland restoration thus requires a habitat-specific analysis of traits that characterize successful colonizers.

Combining floristic, functional and environmental data helps to reveal general characteristics of successful and failed target species (Römermann et al. 2009). While species lists are often region-specific, trait-based characterizations can be more easily transferred across grassland restoration projects. Until now, restoration studies have rarely aimed at relating traits of target species to their establishment success (Pywell et al. 2003; Hedberg & Kotowski 2010) or failure (Hedberg & Kotowski 2010), and only a few studies have related traits to species performance (Roberts et al. 2010). In addition, the relation between traits and environmental settings has rarely been assessed in restoration studies, although it promises to reveal underlying constraints or opportunities.

New potential for establishing grasslands arises in shrinking urban areas. Here, large-scale landscaping methods are needed to green the altered demolition sites that arise among extant apartment buildings (Schetke & Haase 2008). These sites, which can be understood as novel urban ecosystems, are often characterized by high levels of human-mediated disturbance and a range of non-native species that are adapted to novel abiotic settings (Kowarik 2011). As described by Hobbs et al. (2006) for novel (non-urban) ecosystems, novel species assemblages can change the competitive pressure on colonizing species compared to the habitats they usually inhabit. For anthropogenic habitats, Prach & Pyšek (1999) demonstrated the importance of competitive ability and plant height during different successional stages. Competition is thus a specific challenge for native grassland species used in urban restoration projects, as many of the resident species are ruderal species that are often classified as competitive and have high stress and disturbance tolerance (Craine 2005; Grime 2007).

In addition to biotic differences, human-influenced habitats differ greatly from habitats in (semi-)natural surroundings in terms of their abiotic structures (Hobbs et al. 2006). On urban demolition sites, soil features differ from grassland restoration sites of the cultural landscape, such as remnant agricultural fields (e.g. urban sites are likely to have a high stone content) and human-mediated impacts differ from those of agricultural regimes. As is the case for grassland remnants in the cultural landscape (Stein et al. 2008), functional niches in urban settings may remain unfilled because of the absence of diaspore pools in the surroundings (Schleicher et al. 2011). Also, soil seed banks often contain only a few grassland species and more ruderals (Bakker & Berendse 1999); a situation that is likely even more true in highly altered urban soils. The ability of urban wasteland sites to function as novel habitats for grassland species may thus be limited by spatial isolation and missing diaspore pools. This combination of novel abiotic settings, fragmentation and the likely presence of highly competitive non-native or ruderal species poses challenges for grassland restoration on urban sites. To deliberately incorporate such novel settings into habitat management can offer new opportunities to enhance and foster urban biodiversity (Kowarik 2011), but also requires experimentally validated criteria to identify species that can cope with the associated constraints for establishment.

While approaches to enhancing grasslands in the urban context by adding species have been described (Luscombe & Scott 1994; Hitchmough & Woudstra 1999), little evidence on the success of urban grassland establishment and underlying mechanisms exists, and scientifically based experiments that consider initial competition from the resident vegetation are missing thus far. We therefore established an in-situ experiment over 3 yr to test different methods of grassland restoration on urban sites which are isolated from typical grassland sites but neighbour other urban green spaces.

We combined tilling of the soil with sowing seeds of regional provenance to expand opportunities for nature conservation objectives in large-scale urban landscaping. Using native instead of exotic species decreases invasion risks at the species level, and using native species of regional provenance helps to conserve genetic diversity and reduces invasion risks at the sub-species level (Hufford & Mazer 2003). We worked with the given urban soils without adding nutrients or topsoil but tested a mycorrhizal inoculation in one treatment. We deliberately incorporated spontaneous vegetation by activating the soil seed bank. In the third year after treatment, we analysed the performance of target species, species arising from the soil seed bank, species immigrating from the surroundings and resident species from the control plots.

We use trait analyses, which are often performed to explain patterns in species composition (Petchey & Gaston 2006), to shed light on mechanisms that underlie success and failure of experimentally added grassland species. Functional plant traits differ between rural and urban floras (Knapp et al. 2010). Also, trait composition in grassland species changes with increasing urbanization (Williams et al. 2005). We thus hypothesize that success or failure of grassland species sown at urban sites could be related to different functional traits that enable species to cope with the special challenges of urban habitats.

With our study, we therefore attempt to answer the following questions in the context of urban grassland restoration: (1) which plant traits characterize (a) successful target grassland species, (b) failed target grassland species, (c) species arising from the soil seed bank, and (d) species immigrating from the surroundings, and do they differ from those of resident species; (2) how do plant traits of the above-ground vegetation relate to environmental parameters?

Methods

Study area

Situated in northeast Berlin, Germany, Mahrzahn-Hellersdorf is one of many large-scale residential housing areas built in the 1980s. Subsequent severe demographic changes caused an over-supply of apartments and other infrastructure that were demolished in the last decade as a part of the federal programme ‘Stadtumbau Ost’ (Bezirksamt Marzahn-Hellersdorf von Berlin 2007). Shrinkage and perforation of the neighbourhood is obvious: about 100 ha of additional free space – scattered across the district and adjacent to the remaining apartment buildings – exists now, with no anticipated future use (Overmeyer & Renker 2005).

Restoration measures

Within this setting, we set up 11 study sites in autumn 2008. On each site we applied two different grassland restoration treatments in plots of 4 m × 4 m and established an untreated control in a randomized block design. Experimental sites were only tilled prior to treatments, thus diaspores of the resident vegetation were left in the soil. In both treatments, an identical mixture of 27 target species of regional provenances was sown, with species typical of the cultural landscape surrounding Berlin (Table 1). In addition, one of the treatments included a mycorrhizal inoculation. The seeds were provided by commercial producers (Rieger-Hofmann, Blaufelden-Raboldshausen; Saaten-Zeller, Riedern, DE) that collect and propagate seeds of regional provenances. In addition to typical grassland species, we included three rare and endangered species and also, for aesthetic reasons, added the ruderals Papaver rhoeas and Malva alcea, which have attractive flowers. The majority of the target species were not present at the sites before treatments.

Table 1. Data on target species: seed weight, the proportion of in-situ plots with emerging plants, the proportion of plots with more than ten individuals of a target species, and their numbers in the third year after treatment. Successful target species are those occurring in more than four plots per treatment or in ≥14% of plots with more than ten individuals; other species are failed target species
Seed weight per plot (g)% plots with species% plots with ≥10 individualsIndividuals per plot, mean ± SD
Successful target species
 Centaurea jacea subsp. jacea 1.76914112.6 ± 13.7
 Centaurea scabiosa 5.58693.0 ± 2.9
 Dianthus carthusianorum 0.45824112.4 ± 13.9
 Galium mollugo s. str.0.4915017.3 ± 17.7
 Holcus lanatus 0.34954517.3 ± 21.3
 Knautia arvensis s. str.4.255143.7 ± 6.3
 Leucanthemum ircutianum 0.26959139.9 ± 34.8
 Papaver rhoeas 0.095593.9 ± 9.8
 Prunella vulgaris 0.5141142.7 ± 4.8
 Ranunculus acris 1.368278.2 ± 12.1
 Silene vulgaris 0.53955010.5 ± 7.6
Failed target species
 Agrostis capillaris 0.05000.0 ± 0.0
 Armeria maritima subsp. elongata 1.333600.8 ± 1.5
 Campanula rotundifolia 0.025000.0 ± 0.0
 Helichrysum arenarium 0.061800.3 ± 0.7
 Jasione montana 0.121851.1 ± 2.8
 Malva alcea 2.83601.3 ± 2.3
 Plantago media 0.27500.0 ± 0.2
 Rumex acetosella 0.483251.0 ± 2.3
 Silene dioica 0.684593.1 ± 6.0
 Lychnis flos-cuculi 0.15000.0 ± 0.0
 Trifolium arvense 0.38900.1 ± 0.3
Removed from analyses
 Achillea millefolium 0.141009163.0 ± 63.6
 Festuca brevipila 0.5231410.6 ± 34.3
 Festuca ovina 0.32753.1 ± 9.9
 Hypericum perforatum subsp. perforatum 0.084101.0 ± 2.0
 Hypochaeris radicata 0.524602.0 ± 3.1

Prior to seeding, the germination rate of all available target species was tested in a greenhouse under standard conditions (20–25 °C, even light incidence, diaspore-free soil, kept at even moisture, August to October 2008). The number of seeds of each species in the mixture was adjusted according to the germination rate of the greenhouse test so that we could expect 1000 seeds·m−2 to potentially germinate. In the mycorrhizal treatment, we applied 750 ml of the arbuscular mycorrhizal fungus Glomus intraradices (BioMyc Vital , BioMyc Environment GmbH, Brandenburg/Havel, DE) per plot before sowing to determine if mycorrhizal inoculum facilitated vegetation establishment, e.g. by improving soil conditions (Rillig & Mummey 2006) or supporting seedlings (van der Heijden 2004). The plots were mown each autumn.

Data collection

In a 3 m × 3 m relevé centred in each plot, we mapped all target and spontaneous vegetation in the following 2 yr in spring and early and late summer and counted the individuals of each species. For clonally growing species, we assessed ramets with flowers or seeds as separate counting units. Botanical nomenclature was derived from Rothmaler (2005). Target species are the sown species, following the definition of Bakker & Berendse (1999).

Soil seed bank sampling

The soil seed bank sampling followed the standardized method of Thompson et al. (1997). We took a 1-L mixed core sample of ten evenly distributed points from all undisturbed control plots and from all other plots after tilling but before applying the treatments. The samples were stored without light at 1–4 °C for 5 mo. Then, the samples were sieved, concentrated and evenly distributed on diaspore-free soil and germinated in gauze cages in a greenhouse for 1 yr, starting in March 2009. After 6 mo, we slightly disturbed the samples to activate potentially underlying diaspores. Specimens that could not be determined as seedlings were potted and cultivated until determination was possible. Juncus effusus emerged after 9 mo in the control trays, which had been placed between the trays in all cages, and in the trays with soil seed bank samples. We removed this species completely from the data set.

Species data and assignment to groups

The data set for the following analyses includes the individuals present in the plots in the third year after treatment and arising from the soil seed bank samples. Woody species were excluded from the analyses. We assigned species to one of six groups describing their mode of entry to the community and their establishment: (1) ‘successful target species’ were sown target species that were present at the end of the observation period in at least four plots per treatment or in a total of three plots with at least ten individuals per plot; (2) ‘failed target species’ emerged in fewer than four plots per treatment and had fewer than three plots with ten or more individuals; (3) ‘resident species’ were those also present in untreated control plots; (4) ‘successful soil seed bank species’ were spontaneous species found both in the soil seed bank and the above-ground vegetation of treated plots but not in control plots; (5) ‘failed soil seed bank species’ were species found exclusively in the soil seed bank; and (6) ‘immigrating species’ were species that were absent in the soil seed bank and control plots but emerged spontaneously in the treated plots. Five target species (Achillea millefolium, Festuca brevipila, F. ovina, Hypericum perforatum subsp. perforatum, Hypochaeris radicata) were removed from the data set as it could not be determined if their occurrence resulted from immigration, activation of the soil seed bank or seeding. Two target species, Centaurea jacea and Galium mollugo s. str., were also found in a single control plot. However, as they occurred in more than 90% of the treated plots, we left them in the data set of successful target species. We also determined whether a species was associated with the formation type grassland (Haeupler & Muer 2000).

Plant traits

Most plant trait data were derived from the LEDA Traitbase (Kleyer et al. 2008) and the BIOLFLOR database for German flora (Otto 2002) and a few individual traits from Römermann et al. (2005) (Cardamine pratensis agg.) and Moravcová et al. (2010) (Erigeron annuus, Helianthus tuberosus). To describe the competitive ability of grassland species, we chose plant height (m), specific leaf area (m2kg−1) and plant strategies corresponding to the scheme of Grime (1979). To characterize regeneration, we analysed plant life span and seed bank longevity; dispersal traits were assessed as seed mass (mg) and seed shape index (length:width ratio). For most species, the LEDA Traitbase contains numerous entries, so we used the mean of the numeric values for the traits plant height and specific leaf area. For the traits seed mass and seed shape index, we calculated the mean solely from values for the diaspore type ‘germinule’ to gain coherent values of the different databases. For seed bank longevity, we calculated the seed longevity index SLI (Bekker et al. 1998) from at least five records of the database on soil seed banks of North-West Europe (Thompson et al. 1997). Plant life span included the categories ‘annuals’ (annuals, summer annuals, winter annuals), ‘perennials’ (perennials, short-lived perennials, medium-lived perennials) and ‘biennials’ (strict monocarpic biennials, poly-annuals), and plant strategies the categories ‘competitors’ (C-strategists), ‘stress-tolerators’ (S-strategists) and ‘ruderals’ (R-strategists). As one species can belong to several categories (Grime 1979), we generated dummy variables with the categories C, S, R and yes/no entries for each species.

Environmental variables

We assessed various environmental variables to determine the relation between the distribution of plant traits and environmental conditions (Appendix S1). We pooled related initial predictors from PCA analyses that yielded two distinct groups of variables: one group described urban soil features and included the variables soil depth, weight and fraction of stones, oven-dry density, root depth, C/N ratio, pH of upper soil layer, electrical conductivity, phosphorus and potassium. The second group of variables was related to human-mediated impacts and included the variables distance to the next pathway, presence of a fence, mean frequency of people per site, disturbance of soil/vegetation by dogs digging, dog waste, carbonate and bare soil. We used the axis scores of each group of variables for further analyses and henceforth refer to the groups as urban soil features and human-mediated impacts. We additionally considered two variables describing the surrounding vegetation, influence of trees and total vegetation cover (see Table S1). We considered the above-described treatment types with the variable mycorrhizal inoculation (yes/no).

Statistical analyses

To test for differences in traits (1) between successful or failed target and resident species and (2) between successful soil seed bank species, failed soil seed bank species, new immigrating species and resident species, we compared the metric traits of each group to that of resident species with the resampling method of bootstrapping (Crawley 2007; Manly 2007). Although species numbers differed among groups, this method allows direct comparisons (Roff 2006; Köhler et al. 2007). For that, we calculated the means of each trait for the group of successful target species and failed target species and compared them to the corresponding resampled means of the group of resident species. For the latter, out of the original trait set, a random mean was generated with as many random selections as the trait set had entries for resident species. These randomly generated means were resampled 3000 times. The distribution of the resampled means was then compared to the means of the original trait set of the other species groups to detect significant differences (Kühn & Klotz 2006). We provided error probabilities with and without Bonferroni correction to account for multiple comparisons on the same data set. We used a two-tailed Fisher's t-test to determine differences in the categorical traits between the same species groups.

To test for joint effects of plant traits and their interactions on establishment success of target species, we assessed their mean establishment rates in the plots as response in a regression tree model using species traits as predictors. Hereby, the species set is repeatedly split in two groups at a threshold value of a trait that shows a distinctive response to the species’ establishment rate. We pruned the tree model at a maximum difference in risk of one standard error to avoid overfitting of the model and validated it by a ten-fold cross-validation (De'ath & Fabricius 2000). Plant life span was excluded from the predictors, as all but three target species belonged to perennials.

We conducted a fourth-corner analysis (Dray & Legendre 2008) for all species to connect environmental variables and plant traits. A combination of permutation tests was used to relate environmental (plot-based) and trait (species-based) data in terms of the community data for the total above-ground vegetation in our 22 plots. For the community data we used the number of individuals of each species that had established after 3 yr. As data on SLI were not available for 44 species, we did not include this trait in this analysis. The complete trait information (excluding SLI) was available for 120 out of 132 species. We added a variable to assign each species to its species group (see above) to test if the mode of entry to the community (sown/soil seed bank/immigrating) or performance (failed/successful) of a species was related to any environmental variable. We accounted for phylogenetic relatedness of the species by including the genus of each species as a factor in the fourth-corner analysis.

We combined the permutation model types 2 and 4, each repeated 1000 times. We conducted a Monte Carlo test implemented in the fourth-corner analysis (Dray & Legendre 2008) with 1000 permutations to test the assigned relations for significance. To determine the explained variance of the model, we considered the first two RLQ axes. All statistical analyses were conducted with R (version 2.10.1, R Foundation for Statistical Computing, Vienna, AT). For the fourth-corner analysis, the package ade4 was used.

Results

Traits of target species and resident species

Successful target species had similar height and seed mass compared to resident species, while failed target species were significantly smaller and had a lower seed weight (Table 2, Fig. 1a). Compared to resident species, the specific leaf area was smaller for successful target species and similar for failed target species (Table 2, Fig. 1b). The SLI was significantly smaller for successful target species and similar for failed target compared to resident species (Table 2). Both successful and failed target species had smaller seed shape indices than resident species (Table 2). In the third year after treatment, most established species were perennials in all groups (Table 3). There were no significant differences between species groups for ecological strategy types (Table 3).

Figure 1.

Differences in plant traits of successful and failed target species compared to resident species of urban wasteland sites for (a) plant height and (b) specific leaf area. Broken lines show the 95% confidence interval of the resampled means. Species groups that fall within the lines are similar to resident species; outside these lines, they have either a significantly smaller (−) or a significantly larger (+) mean value compared to resident species.

Table 2. Results of resampling five plant traits of successful and failed target species and comparison to resident species. n, number of species with available trait data and used for calculation; P adj, Bonferroni-corrected error probability. Significant differences of the species group compared to resident species are in bold. For ease of interpretation, ⇓ and ⇑ indicate if the mean of the given trait is significantly lower or higher compared to resident species
n Mean ± SDResampled mean ± SD, relation P P adj
Plant height (m)
Successful target species110.42 ± 0.23 0.681.38
Failed target species110.27 ± 0.18 <0.001 <0.001
Resident species76 0.43 ± 0.03
Specific leaf area
Successful target species1122.95 ± 6.91 <0.001 <0.01
Failed target species1124.84 ± 8.57 0.320.652
Resident species74 25.63 ± 0.81
Seed mass (mg)
Successful target species111.73 ± 2.13 0.250.50
Failed target species110.62 ± 1.11 <0.01 <0.05
Resident species75 2.60 ± 0.75
Seed shape index (length:width)
Successful target species111.94 ± 0.70 <0.01 <0.01
Failed target species111.97 ± 0.79 <0.01 <0.01
Resident species75 2.82 ± 0.29
Seed longevity index (SLI)
Successful target species100.43 ± 0.27 <0.001 <0.001
Failed target species70.57 ± 0.14 0.601.21
Resident species50 0.56 ± 0.03
Table 3. Representation of the categorical traits ‘plant life span’ and ‘ecological strategy type’ (Grime 1979) for the different species groups. Differences were tested with Fisher's test, with each species group compared to resident species. Significant differences between species groups are in bold
Species groupGrassland species (%)Plant life span (%)Ecological strategy type (%)
AnnualBiennialPerennial P CompetitorsStress toleratorsRuderals P
Successful target species (n = 11)829091 0.01 5520250.40
Failed target species (n = 11)8201882 <0.01 3831310.20
Successful soil seed bank species (n = 12)503325420.304118410.95
Failed soil seed bank species (n = 36)58448471.04218400.88
Immigrating species (n = 20) 80 14581 0.01 4231280.10
Resident species (n = 76)6845946 431641

Traits of spontaneous species

The comparison of soil seed bank and immigrating species to resident species in the above-ground vegetation showed that soil seed bank species that succeeded in establishment were taller, whereas immigrating species were shorter than resident species (Table 4, Fig. 2a). Failed soil seed bank species were similar to resident species in height. All groups of spontaneous, i.e. non-sown, species were similar in their specific leaf area and seed mass. The seed shape index was smaller for successful soil seed bank species, larger for immigrating species and similar for failed soil seed bank species when compared to the group of resident species. SLI was higher in successful soil seed bank species and smaller in immigrating species (Fig. 2b).

Figure 2.

Differences in plant traits of successful and failed soil seed bank species (SSB) and immigrating species compared to resident species at urban wasteland sites for (a) plant height and (b) seed longevity index (SLI). Broken lines show the 95% confidence interval of the resampled means. Species groups that fall within the lines are similar to resident species; outside these lines, they have either a significantly smaller (−) or a significantly larger (+) mean value compared to resident species.

Table 4. Results of resampling five plant traits of different groups of soil seed bank and immigrating species compared to resident species. n, number of species with available trait data and used for calculation; P adj, Bonferroni-corrected error probability. Significant differences of the species group compared to resident species are in bold. For ease of interpretation, ⇓ and ⇑ indicate if the mean is significantly lower or higher compared to resident species
n Mean ± SDResampled mean ± SD, relation P P adj
Plant height (m)
Successful soil seed bank species120.53 ± 0.48 <0.01 <0.05
Failed soil seed bank species360.39 ± 0.25 0.210.66
Immigrating species210.36 ± 0.22 <0.05 0.10
Resident species76 0.43 ± 0.03
Specific leaf area
Successful soil seed bank species1225.97 ± 7.70 0.702.12
Failed soil seed bank species3626.81 ± 10.03 0.160.48
Immigrating species2124.87 ± 10.87 0.320.97
Resident species74 25.67 ± 0.81
Seed mass (mg)
Successful soil seed bank species121.19 ± 1.63 0.060.20
Failed soil seed bank species361.35 ± 4.57 0.100.31
Immigrating species203.53 ± 6.68 0.200.63
Resident species75 2.58 ± 0.75
Seed shape index (length:width)
Successful soil seed bank species121.81 ± 0.70 <0.001 <0.001
Failed soil seed bank species272.72 ± 2.60 0.722.18
Immigrating species203.93 ± 4.17 <0.001 <0.001
Resident species75 2.83 ± 0.29
Seed longevity index (SLI)
Successful soil seed bank species80.70 ± 0.25 <0.001 <0.001
Failed soil seed bank species230.61 ± 0.27 0.060.20
Immigrating species120.49 ± 0.28 <0.05 0.06
Resident species50 0.56 ± 0.03

Among the soil seed bank and immigrating species that established in the third year after treatment, perennials prevailed (Table 3). The proportion of competitors was similar in all groups (ca. 40%), and no significant differences in ecological strategy were determined among species groups (Table 3).

Joint effects of traits on target species success

The pruned regression tree model (Fig. 3) showed an interaction of two traits, whereas all other traits were eliminated from the model as they increased cross-validated errors. The most important predictor for the mean establishment rate of target species in a plot was plant height, predicting a low establishment rate (0.177) for species shorter than 0.255 m and a high establishment rate for taller plants (0.689). Only the group of taller species was further divided by seed mass (threshold of 0.244 g). Heavier seeds had the highest establishment rate (0.764), in contrast with lighter seeds, which had an intermediate rate (0.273).

Figure 3.

Regression tree showing that two traits, plant height and seed mass, explain the establishment rate of target species in the plots. The group of taller plants is split up into those target species with smaller and larger seeds. Numbers below the nodes represent the mean establishment success in this (sub)group; n, number of target species per node.

Role of environmental settings in relation to traits

In the fourth-corner analysis, the species' mode of entry to the community and genus did not significantly relate to any environmental variables nor was the treatment type related to any traits (Table 5). That is, the relationships between above-ground vegetation, traits and environmental variables were independent of species group, phylogenetic composition of the species assemblage and mycorrhizal treatment type. The environmental variables human-mediated impact and influence of trees were both positively related to the trait variable plant life form annual (Fig. 4b) and negatively to the plant life form perennial. The strategy type competitor was negatively (Fig. 4c) and specific leaf area (Fig. 4a) was positively correlated to both human-mediated impact and influence of trees. With a P-value < 0.001, the model has an explained variance of 69% on the first RLQ axis and 14% on the second RLQ axis.

Figure 4.

Human-mediated impacts shaped the distributions of functional plant traits at the treated urban wasteland sites. Shown is the relation of the traits (a) specific leaf area, (b) plant life form (annuals), (c) strategy type (competitors) to the environmental variable human-mediated impacts. Significant relations between environmental variables and traits were derived by fourth-corner analysis (Table 5).

Table 5. Results of fourth-corner analysis showing all of the significant relations between environmental and trait variables; correlations between plant traits and environmental variables and corresponding P-values are shown (for plant life form and strategy type, P-values are adjusted with Holmes correction). Traits that did not show a significant relationship were plant height, seed mass, seed shape index, SLI, species group and genus. Environmental variables that did not show a significant relationship were urban soil features, total vegetation cover and treatment type
Trait variableEnvironmental variable
Human-mediated impactInfluence of trees
Correlation P Correlation P
Plant life form: annuals0.45<0.010.40<0.01
Strategy type: competitors−0.37<0.01−0.31<0.05
Specific leaf area0.26<0.050.290.01

Discussion

In restoration projects, information on the relation between plant traits and establishment success is expected to enhance knowledge transfer independently of regional settings and local species pools. Relating traits of successful and failed target species in addition to information about the environmental setting broadens knowledge for restoration practice (Pywell et al. 2003; Hedberg & Kotowski 2010; Roberts et al. 2010).

In the context of urban grassland restoration, underlying mechanisms of colonizing species might differ from those of grassland restoration sites in rural landscapes as competition with ruderal and non-native species in urban sites is likely more pronounced and the environmental conditions of the different settings often contrast. For grassland restoration of urban demolition sites, our results revealed that successful and failed target species, species recruiting from the soil seed bank and species immigrating from the surroundings had traits that were clearly distinguishable from those of the resident species.

Architectural traits that account for competitive ability of target species

In our study, plant height was a main characteristic of successful target species in urban wasteland sites 3 yr after treatment (Table 2, Fig. 1a). As this parameter is often associated with competitive vigour (Cornelissen et al. 2003), our results suggest that target species should be at least as tall as resident species. Although taller plants usually have a higher investment during early establishment, they later profit from better access to light compared to smaller plants (Westoby et al. 2002). Results from grasslands with experimentally modified species richness show that plant height of colonizing species increases with species richness (Schmidtke et al. 2010). This is likely a direct outcome of an increased vertical density at sites of higher plant species richness (Lorentzen et al. 2008). Thus, an already medium to high richness of the wasteland vegetation in urban demolition sites obviously leaves few vacant vertical niches for target species below the mean plant height of the resident vegetation.

Another result of our study showed the importance of a low specific leaf area of successful target species relative to that of resident species (Table 2, Fig. 1b). Small specific leaf area generally enables plants to deal with high light incidence or resource-limited settings, such as those with low water availability (Cornelissen et al. 2003; Cousins & Lindborg 2004). Both conditions were true for our open demolition sites, where stony soils prevail, similar to other urban wastelands (Godefroid et al. 2007). Additionally, leaf longevity and structural strength of the leaves indicated by a low specific leaf area (Funk et al. 2008) might be advantageous in urban settings where physical disturbance by humans or dogs is common.

Seed traits as determinants of establishment

Seed mass is a key trait in grassland colonization, as species with larger seed mass have larger seedlings due to greater reserves (Cornelissen et al. 2003) and germinate best in mown grassland plots (Kahmen & Poschlod 2008). Species with smaller seed mass are thus less competitive while establishing in resident vegetation and competing with larger seedlings. The main advantage of small seeds – better dispersal ability – is negligible when species are sown. This is consistent with our results, where the mean seed mass of successful target species was similar to that of resident species whereas that of failed target species was smaller than that of resident species (Table 2). Hence, similar to our results for plant height (Table 2), for seed mass, it is the relation of the trait state of sown to resident species that is of importance. It is thus likely that competition between seedlings of sown and resident species during the recovery of the tilled wasteland vegetation was a crucial process in separating successful and failed target species. This is supported by the regression tree model, showing that the mean establishment rate of a target species is determined by its height, and that the group of larger plants separates into those with larger or smaller seeds (Fig. 3). The order of nodes in the tree model also suggests that effects of plant height dominate over other traits, but that seed mass is particularly important for discriminating the success of taller target species.

Our analysis clearly suggests that seed shape index does not matter for the establishment success of target species, as it was smaller for both successful and failed target species compared to resident species (Table 2).

Plant strategies

One surprising result of our study was the absence of differences in CSR strategy types among species groups (Table 3). Instead, differences in plant height – an architectural trait associated with competitive ability – were significantly related to establishment success (see above). Nevertheless, the importance of strategy type can change over time, as Pywell et al. (2002) reported that ruderality (R-component) is an important strategy in the first year of establishment of grassland forbs, while stress tolerance (S-strategists) and competitiveness (C-strategists) become more important later.

Characteristics of soil seed bank and immigrating species

Fifty per cent of successful soil seed bank species and 80% of immigrating species were grassland species (Table 3). This suggests a potential of non-sown grassland species to establish after the resident vegetation and soil are disturbed through tilling. Similarly to successful target species, in soil seed bank species, plant height was a main characteristic of successful establishment on urban wasteland sites 3 yr after treatment (Table 4, Fig. 2a). Although many short-lived species are found in urban species assemblages (Sudnik-Wójcikowska & Galera 2005; Knapp et al. 2008a), successful soil seed bank species and immigrating species had high proportions of perennials (Table 3). In terms of seed shape index, successful soil bank species tended to be more spherical (smaller index) than resident species (Table 4), whereas immigrating species were less spherical. For all species, this suggests that seed shape index did not limit establishment success once the diaspore entered the site. For species of the soil seed bank, this suggests, however, that they are likely able to persist longer in the soil due to their more spherical shape (Weiher et al. 1999; Lavorel & Garnier 2002) and their considerably high SLI (Table 4). This allows these species to germinate after remaining undisturbed for long periods.

In rural restoration projects, only a few grassland target species are found within soil seed banks, and restoration success differs greatly among grassland types and sites, even if they are geographically close (Latzel et al. 2011; Valkó et al. 2011). Activation of soil seed bank species alone is thus not sufficient to create typical meadows (Nordbakken et al. 2010). Few grassland target species were found in the soil seed bank of sites subjected to long-term cutting regimes, even though they were found growing in the above-ground vegetation (Bakker et al. 2002). Generally speaking, grassland species do not invest in long-lived seed banks, but in the competitive potential of seedlings (Hedberg & Kotowski 2010). This suggests a low potential for grassland species in rural settings to establish from the soil seed bank. In urban wasteland sites, however, species likely differ in the composition of traits that reflect regeneration potential. For example, on urban wasteland sites, a persistent soil seed bank was found to be advantageous for coping with high fragmentation (Westermann et al. 2011).

Although additional grassland species did establish from the soil seed bank on our urban demolition sites, immigration was the more common mode of entry for grassland species to the experimental plots (80% of 21 immigrating species were grassland species vs 50% of 13 soil seed bank species; Table 3). The lower SLI and the higher seed shape index of immigrating species (Table 4) suggest that this species group is less persistent than spontaneous species of the soil seed bank and resident species. Latzel et al. (2011) showed consistently that more species established through other modes of entry than by the activation of the soil seed bank. Human-mediated dispersal increasingly matters in grassland species (Auffret 2011). Vehicles, for example, move seeds of grassland species in urban settings (von der Lippe & Kowarik 2008), but the spatial reach of this dispersal pathway beyond roadside habitats to fragmented urban wastelands remains to be studied.

Traits related to urban pressure

In the fourth-corner analysis, the first RLQ axis offered a good explanatory value, explaining 69% of the variance (Table 5). Increasing human-mediated impact was related to higher specific leaf area, a higher proportion of annuals and a lower proportion of competitors. Hence, as human-mediated impact increases, species tend to have softer leaves, a shorter life cycle and follow stress-tolerant or ruderal strategies. Our finding of more annuals in areas with increasing disturbance as shown here at the population level partially matches results found at larger scales (Knapp et al. 2008b).

Low specific leaf area is a characteristic by which successful target species differed from the resident vegetation (Table 2). We assume that successful targets here occupy a vacant niche, that is, they grow rather slowly while efficiently using existing resources and rely on their resistance to drought. This ability can be significant in stony, urban soils although the trait was unrelated to urban soil conditions. However, to fill this vacant niche, successful target species need to be tall enough and their diaspores need to have sufficient resources for germination, otherwise they will be overgrown by the regenerating resident vegetation.

Regarding the whole species assemblage, nonetheless, a high specific leaf area is generally a characteristic that allows plants to deal with human pressure, probably due to an enhanced ability to regenerate after disturbance. It remains to be seen whether target species can establish over longer periods in plots with high human pressure, or if their lower specific leaf area is a limit to their success. At the same time, human pressure itself may be the cause of this empty niche, as generally few species of the diaspore pool may be able to cope with it.

As determined with the fourth-corner analysis, the combination of species and environmental setting in our study sites was not related to the phylogenetic structure of the species assemblage (Table 4). Also, the mycorrhizal treatment did not influence trait composition within the species assemblages, nor were environmental variables related to the mode of species' entry to the community (Table 5). That is, trait–environmental relations were not influenced by whether species originated from sowing, from the soil seed bank or from surrounding diaspore pools.

Conclusions for grassland restoration in urban settings

While grassland species are strongly declining in traditional cultural landscapes (Henle et al. 2008), urban sites harbour potential habitats for them (Maurer et al. 2000; DeCandido et al. 2007), although novel urban conditions (Kowarik 2011) involve many possible constraints for grassland restoration. Our experiment in grassland establishment on urban wasteland sites may be perceived as an ecological intervention (Hobbs et al. 2011) rather than the restoration of typical grassland assemblages.

In this study we revealed a set of plant traits that clearly enabled reintroduced grassland species to cope with specific urban challenges. Successful target species can either mimic competitive traits of the resident species (e.g. similar plant height and seed mass) or differ from them (e.g. lower specific leaf area). Grassland species that can successfully be integrated into urban wasteland vegetation may thus fill well-defined vacant niches, while resembling the traits of the resident species in other ways. The lack of appropriate diaspore input within urban fragmented landscapes may allow for vacant niches in wasteland vegetation, as even common ruderal species are dispersal limited (Schleicher et al. 2011). As our trait-based characterization of successful target species relies on relative comparisons with the spontaneous species, we argue that it is transferable to similar urban settings with different resident vegetation.

Activation of the soil seed bank by tilling offers an additional opportunity for the establishment of grassland species, but here establishment success was associated with different traits compared to sown target species. Successful soil seed bank species, which included 50% grassland species, were taller than resident species and had seed traits that allowed for longer persistence in the soil (smaller seed shape index and higher SLI compared to resident species). The overall relation of traits and environmental variables showed that human-induced disturbances tend to select for traits that increase a plant's ability to cope with stress and ruderality (increasing proportion of annuals, less reliance on the competitor strategy).

Acknowledgements

This study was funded by the German Research Foundation (DFG) as part of the graduate research training programme ‘Urban Ecology Berlin’ (GRK 780/III). We thank Birgit Seitz for help with species determination, Verena Rodorff and Destin Lau for assistance in the field, Gabriele Hinz and Yoganathan Gopalasamy for help with greenhouse work, Norbert Kühn for advice and his gardeners for practical support, Rüdiger Wittig and two anonymous referees for constructive comments and Kelaine Vargas for improving our English.

Ancillary